Face Generation¶

In this project, you'll use generative adversarial networks to generate new images of faces.

Get the Data¶

You'll be using two datasets in this project:

MNIST
CelebA

Since the celebA dataset is complex and you're doing GANs in a project for the first time, we want you to test your neural network on MNIST before CelebA. Running the GANs on MNIST will allow you to see how well your model trains sooner.

If you're using FloydHub, set data_dir to "/input" and use the FloydHub data ID "R5KrjnANiKVhLWAkpXhNBe".

data_dir = './data'

# FloydHub - Use with data ID "R5KrjnANiKVhLWAkpXhNBe"
#data_dir = '/input'


"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import helper

helper.download_extract('mnist', data_dir)
helper.download_extract('celeba', data_dir)

Downloading mnist: 9.92MB [00:09, 1.06MB/s]                                                                  
Extracting mnist: 100%|████████████████████████████████████████████████| 60.0K/60.0K [01:38<00:00, 606File/s]
Downloading celeba: 1.44GB [13:20, 1.80MB/s]

Extracting celeba...

Explore the Data¶

MNIST¶

As you're aware, the MNIST dataset contains images of handwritten digits. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
%matplotlib inline
import os
from glob import glob
from matplotlib import pyplot

mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'mnist/*.jpg'))[:show_n_images], 28, 28, 'L')
pyplot.imshow(helper.images_square_grid(mnist_images, 'L'), cmap='gray')

<matplotlib.image.AxesImage at 0x1ce8eafa278>

CelebA¶

The CelebFaces Attributes Dataset (CelebA) dataset contains over 200,000 celebrity images with annotations. Since you're going to be generating faces, you won't need the annotations. You can view the first number of examples by changing show_n_images.

show_n_images = 25

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
mnist_images = helper.get_batch(glob(os.path.join(data_dir, 'img_align_celeba/*.jpg'))[:show_n_images], 28, 28, 'RGB')
pyplot.imshow(helper.images_square_grid(mnist_images, 'RGB'))

<matplotlib.image.AxesImage at 0x1cd9f01fac8>

Preprocess the Data¶

Since the project's main focus is on building the GANs, we'll preprocess the data for you. The values of the MNIST and CelebA dataset will be in the range of -0.5 to 0.5 of 28x28 dimensional images. The CelebA images will be cropped to remove parts of the image that don't include a face, then resized down to 28x28.

The MNIST images are black and white images with a single color channel while the CelebA images have 3 color channels (RGB color channel).

Build the Neural Network¶

You'll build the components necessary to build a GANs by implementing the following functions below:

model_inputs
discriminator
generator
model_loss
model_opt
train

Check the Version of TensorFlow and Access to GPU¶

This will check to make sure you have the correct version of TensorFlow and access to a GPU

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
from distutils.version import LooseVersion
import warnings
import tensorflow as tf

# Check TensorFlow Version
assert LooseVersion(tf.__version__) >= LooseVersion('1.0'), 'Please use TensorFlow version 1.0 or newer.  You are using {}'.format(tf.__version__)
print('TensorFlow Version: {}'.format(tf.__version__))

# Check for a GPU
if not tf.test.gpu_device_name():
    warnings.warn('No GPU found. Please use a GPU to train your neural network.')
else:
    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))

TensorFlow Version: 1.0.0
Default GPU Device: /gpu:0

Input¶

Implement the model_inputs function to create TF Placeholders for the Neural Network. It should create the following placeholders:

Real input images placeholder with rank 4 using image_width, image_height, and image_channels.
Z input placeholder with rank 2 using z_dim.
Learning rate placeholder with rank 0.

Return the placeholders in the following the tuple (tensor of real input images, tensor of z data)

import problem_unittests as tests

def model_inputs(image_width, image_height, image_channels, z_dim):
    """
    Create the model inputs
    :param image_width: The input image width
    :param image_height: The input image height
    :param image_channels: The number of image channels
    :param z_dim: The dimension of Z
    :return: Tuple of (tensor of real input images, tensor of z data, learning rate)
    """
    return tf.placeholder(tf.float32, shape=([None, image_width, image_height, image_channels]), name='input_real'), \
                tf.placeholder(tf.float32, shape=[None, z_dim], name='input_z'), \
                tf.placeholder(tf.float32, shape=None, name='learning_rate')


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_inputs(model_inputs)

Tests Passed

Discriminator¶

Implement discriminator to create a discriminator neural network that discriminates on images. This function should be able to reuse the variabes in the neural network. Use tf.variable_scope with a scope name of "discriminator" to allow the variables to be reused. The function should return a tuple of (tensor output of the generator, tensor logits of the generator).

keep_prob = 0.8

def discriminator(images, reuse=False):
    """
    Create the discriminator network
    :param image: Tensor of input image(s)
    :param reuse: Boolean if the weights should be reused
    :return: Tuple of (tensor output of the discriminator, tensor logits of the discriminator)
    """
    # TODO: Implement Function
    with tf.variable_scope('discriminator', reuse=reuse):       
        # Input layer is 28x28x3
        # https://prateekvjoshi.com/2016/03/29/understanding-xavier-initialization-in-deep-neural-networks/
        x1 = tf.layers.conv2d(images, 64, 5, strides=2, padding='same', kernel_initializer=tf.contrib.layers.xavier_initializer())
        relu1 = leaky_relu(x1)
        x2 = tf.layers.conv2d(relu1, 128, 5, strides=2, padding='same', kernel_initializer=tf.contrib.layers.xavier_initializer())
        batch_norm2 = tf.layers.batch_normalization(x2, training=True)
        relu2 = leaky_relu(batch_norm2)
# Discriminator should be smaller model than generator
#         x3 = tf.layers.conv2d(relu2, 256, 5, strides=2, padding='same', kernel_initializer=tf.contrib.layers.xavier_initializer())
#         batch_norm3 = tf.layers.batch_normalization(x3, training=True)
#         relu3 = leaky_relu(batch_norm3)
        #add drop out 
        drop2 = tf.nn.dropout(relu2, keep_prob)

        # Flatten
        flat = tf.reshape(drop2, (-1, 4*4*256))
        logits = tf.layers.dense(flat, 1)
        logits = tf.nn.dropout(logits, keep_prob)
        out = tf.sigmoid(logits)
    return out, logits


def leaky_relu(node, alpha=0.01, name='leaky_relu'):
    return tf.maximum(alpha * node, node, name=name)

"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_discriminator(discriminator, tf)

Tests Passed

Generator¶

Implement generator to generate an image using z. This function should be able to reuse the variabes in the neural network. Use tf.variable_scope with a scope name of "generator" to allow the variables to be reused. The function should return the generated 28 x 28 x out_channel_dim images.

def generator(z, out_channel_dim, is_train=True):
    """
    Create the generator network
    :param z: Input z
    :param out_channel_dim: The number of channels in the output image
    :param is_train: Boolean if generator is being used for training
    :return: The tensor output of the generator
    """
    with tf.variable_scope('generator', reuse= not is_train):

        x1 = tf.layers.dense(z, 7*7*256)
        x1 = tf.nn.dropout(x1, keep_prob) # add drop out
        x1 = tf.reshape(x1, (-1, 7, 7, 256))
        x1 = tf.layers.batch_normalization(x1, training=is_train)
        x1 = leaky_relu(x1)
        
        x2 = tf.layers.conv2d_transpose(x1, 128, 5, strides=2, padding='same')
        x2 = tf.layers.batch_normalization(x2, training=is_train)
        x2 = leaky_relu(x2)

        x3 = tf.layers.conv2d_transpose(x2, 64, 5, strides=2, padding='same')
        x3 = tf.layers.batch_normalization(x3, training=is_train)
        x3 = leaky_relu(x3)
            
        logits = tf.layers.conv2d_transpose(x3, out_channel_dim, 5, strides=1, padding='same')
        out = tf.tanh(logits)    
    return out


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_generator(generator, tf)

Tests Passed

Loss¶

Implement model_loss to build the GANs for training and calculate the loss. The function should return a tuple of (discriminator loss, generator loss). Use the following functions you implemented:

discriminator(images, reuse=False)
generator(z, out_channel_dim, is_train=True)

def model_loss(input_real, input_z, out_channel_dim):
    """
    Get the loss for the discriminator and generator
    :param input_real: Images from the real dataset
    :param input_z: Z input
    :param out_channel_dim: The number of channels in the output image
    :return: A tuple of (discriminator loss, generator loss)
    """
    g_model = generator(input_z, out_channel_dim)
    d_model_real, d_logits_real = discriminator(input_real)
    d_model_fake, d_logits_fake = discriminator(g_model, reuse=True)

    d_loss_real = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_real, labels=tf.ones_like(d_model_real)))
    d_loss_fake = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.zeros_like(d_model_fake)))
    g_loss = tf.reduce_mean(
        tf.nn.sigmoid_cross_entropy_with_logits(logits=d_logits_fake, labels=tf.ones_like(d_model_fake)))

    d_loss = d_loss_real + d_loss_fake

    return d_loss, g_loss


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_loss(model_loss)

Tests Passed

Optimization¶

Implement model_opt to create the optimization operations for the GANs. Use tf.trainable_variables to get all the trainable variables. Filter the variables with names that are in the discriminator and generator scope names. The function should return a tuple of (discriminator training operation, generator training operation).

def model_opt(d_loss, g_loss, learning_rate, beta1):
    """
    Get optimization operations
    :param d_loss: Discriminator loss Tensor
    :param g_loss: Generator loss Tensor
    :param learning_rate: Learning Rate Placeholder
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :return: A tuple of (discriminator training operation, generator training operation)
    """
    # TODO: Implement Function
    t_vars = tf.trainable_variables()
    d_vars = [var for var in t_vars if var.name.startswith('discriminator')]
    g_vars = [var for var in t_vars if var.name.startswith('generator')]

    # Optimize
    d_train_o = tf.train.AdamOptimizer(learning_rate, beta1=beta1).minimize(d_loss, var_list=d_vars)

    update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
    g_updates = [opt for opt in update_ops if opt.name.startswith('generator')]
    with tf.control_dependencies(g_updates):
       g_train_o = tf.train.AdamOptimizer(learning_rate, beta1).minimize(g_loss, var_list=g_vars)    
    
    return d_train_o, g_train_o


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
tests.test_model_opt(model_opt, tf)

Tests Passed

Neural Network Training¶

Show Output¶

Use this function to show the current output of the generator during training. It will help you determine how well the GANs is training.

"""
DON'T MODIFY ANYTHING IN THIS CELL
"""
import numpy as np

def show_generator_output(sess, n_images, input_z, out_channel_dim, image_mode):
    """
    Show example output for the generator
    :param sess: TensorFlow session
    :param n_images: Number of Images to display
    :param input_z: Input Z Tensor
    :param out_channel_dim: The number of channels in the output image
    :param image_mode: The mode to use for images ("RGB" or "L")
    """
    cmap = None if image_mode == 'RGB' else 'gray'
    z_dim = input_z.get_shape().as_list()[-1]
    example_z = np.random.uniform(-1, 1, size=[n_images, z_dim])

    samples = sess.run(
        generator(input_z, out_channel_dim, False),
        feed_dict={input_z: example_z})

    images_grid = helper.images_square_grid(samples, image_mode)
    pyplot.imshow(images_grid, cmap=cmap)
    pyplot.show()

Train¶

Implement train to build and train the GANs. Use the following functions you implemented:

model_inputs(image_width, image_height, image_channels, z_dim)
model_loss(input_real, input_z, out_channel_dim)
model_opt(d_loss, g_loss, learning_rate, beta1)

Use the show_generator_output to show generator output while you train. Running show_generator_output for every batch will drastically increase training time and increase the size of the notebook. It's recommended to print the generator output every 100 batches.

def train(epoch_count, batch_size, z_dim, learning_rate, beta1, get_batches, data_shape, data_image_mode):
    """
    Train the GAN
    :param epoch_count: Number of epochs
    :param batch_size: Batch Size
    :param z_dim: Z dimension
    :param learning_rate: Learning Rate
    :param beta1: The exponential decay rate for the 1st moment in the optimizer
    :param get_batches: Function to get batches
    :param data_shape: Shape of the data
    :param data_image_mode: The image mode to use for images ("RGB" or "L")
    """
    # The inputs
    input_real, input_z, lr = model_inputs(data_shape[1], data_shape[2], data_shape[3], z_dim)
    
    # Loss
    d_loss, g_loss = model_loss(input_real, input_z, data_shape[3])
    
    # Optimization
    d_train_opt, g_train_opt = model_opt(d_loss, g_loss, learning_rate, beta1)
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for epoch_i in range(epoch_count):
            steps = 0
            for batch_images in get_batches(batch_size):
                batch_images = batch_images.reshape(batch_size, data_shape[1], data_shape[2], data_shape[3])
                batch_images = batch_images*2

                z_noise = np.random.uniform(-1, 1, size=(batch_size, z_dim))

                _ = sess.run(d_train_opt, feed_dict={input_real: batch_images, input_z: z_noise})
                _ = sess.run(g_train_opt, feed_dict={input_z: z_noise})
                steps += 1

                train_loss_d = sess.run(d_loss, feed_dict={input_real: batch_images, input_z: z_noise})
                train_loss_g = g_loss.eval({input_z: z_noise})

                if steps % 10 == 0:
                    print("Epoch {}/{}: steps {}".format(epoch_i+1, epochs, steps),
                              "Discriminator Loss: {:.4f}...".format(train_loss_d),
                              "Generator Loss: {:.4f}".format(train_loss_g))

                if steps % 100 == 0:
                    show_generator_output(sess, 10, input_z, data_shape[3], data_image_mode)

MNIST¶

Test your GANs architecture on MNIST. After 2 epochs, the GANs should be able to generate images that look like handwritten digits. Make sure the loss of the generator is lower than the loss of the discriminator or close to 0.

batch_size = 64
z_dim = 100
learning_rate = .0003
beta1 = .5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 5

mnist_dataset = helper.Dataset('mnist', glob(os.path.join(data_dir, 'mnist/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, mnist_dataset.get_batches,
          mnist_dataset.shape, mnist_dataset.image_mode)

Epoch 1/5: steps 10 Discriminator Loss: 1.2844... Generator Loss: 0.6878
Epoch 1/5: steps 20 Discriminator Loss: 1.3078... Generator Loss: 0.8317
Epoch 1/5: steps 30 Discriminator Loss: 0.8799... Generator Loss: 1.2109
Epoch 1/5: steps 40 Discriminator Loss: 1.1637... Generator Loss: 0.7993
Epoch 1/5: steps 50 Discriminator Loss: 0.9290... Generator Loss: 1.0208
Epoch 1/5: steps 60 Discriminator Loss: 0.9290... Generator Loss: 1.1219
Epoch 1/5: steps 70 Discriminator Loss: 1.3494... Generator Loss: 0.8403
Epoch 1/5: steps 80 Discriminator Loss: 1.3122... Generator Loss: 0.7262
Epoch 1/5: steps 90 Discriminator Loss: 1.2294... Generator Loss: 0.7808
Epoch 1/5: steps 100 Discriminator Loss: 1.2077... Generator Loss: 0.8382

Epoch 1/5: steps 110 Discriminator Loss: 1.2874... Generator Loss: 0.8538
Epoch 1/5: steps 120 Discriminator Loss: 1.1600... Generator Loss: 0.7594
Epoch 1/5: steps 130 Discriminator Loss: 1.2797... Generator Loss: 0.7129
Epoch 1/5: steps 140 Discriminator Loss: 1.2119... Generator Loss: 0.7944
Epoch 1/5: steps 150 Discriminator Loss: 1.1741... Generator Loss: 0.9199
Epoch 1/5: steps 160 Discriminator Loss: 1.2822... Generator Loss: 0.8651
Epoch 1/5: steps 170 Discriminator Loss: 1.2700... Generator Loss: 0.8286
Epoch 1/5: steps 180 Discriminator Loss: 1.1507... Generator Loss: 0.9799
Epoch 1/5: steps 190 Discriminator Loss: 1.2797... Generator Loss: 0.8113
Epoch 1/5: steps 200 Discriminator Loss: 1.2025... Generator Loss: 0.8467

Epoch 1/5: steps 210 Discriminator Loss: 1.2619... Generator Loss: 0.8831
Epoch 1/5: steps 220 Discriminator Loss: 1.3465... Generator Loss: 0.7986
Epoch 1/5: steps 230 Discriminator Loss: 1.3250... Generator Loss: 0.8387
Epoch 1/5: steps 240 Discriminator Loss: 1.2980... Generator Loss: 0.8028
Epoch 1/5: steps 250 Discriminator Loss: 1.2577... Generator Loss: 0.7646
Epoch 1/5: steps 260 Discriminator Loss: 1.2713... Generator Loss: 0.9242
Epoch 1/5: steps 270 Discriminator Loss: 1.3119... Generator Loss: 0.7220
Epoch 1/5: steps 280 Discriminator Loss: 1.2589... Generator Loss: 0.7767
Epoch 1/5: steps 290 Discriminator Loss: 1.2631... Generator Loss: 0.8329
Epoch 1/5: steps 300 Discriminator Loss: 1.2358... Generator Loss: 0.8531

Epoch 1/5: steps 310 Discriminator Loss: 1.2274... Generator Loss: 0.7437
Epoch 1/5: steps 320 Discriminator Loss: 1.2999... Generator Loss: 0.7945
Epoch 1/5: steps 330 Discriminator Loss: 1.2673... Generator Loss: 0.9083
Epoch 1/5: steps 340 Discriminator Loss: 1.3479... Generator Loss: 0.7933
Epoch 1/5: steps 350 Discriminator Loss: 1.2532... Generator Loss: 0.9143
Epoch 1/5: steps 360 Discriminator Loss: 1.2490... Generator Loss: 0.9042
Epoch 1/5: steps 370 Discriminator Loss: 1.3252... Generator Loss: 0.8447
Epoch 1/5: steps 380 Discriminator Loss: 1.1634... Generator Loss: 0.8080
Epoch 1/5: steps 390 Discriminator Loss: 1.3041... Generator Loss: 0.7611
Epoch 1/5: steps 400 Discriminator Loss: 1.2345... Generator Loss: 0.8441

Epoch 1/5: steps 410 Discriminator Loss: 1.2543... Generator Loss: 0.7493
Epoch 1/5: steps 420 Discriminator Loss: 1.2539... Generator Loss: 0.8263
Epoch 1/5: steps 430 Discriminator Loss: 1.3030... Generator Loss: 0.8390
Epoch 1/5: steps 440 Discriminator Loss: 1.2808... Generator Loss: 0.8310
Epoch 1/5: steps 450 Discriminator Loss: 1.1767... Generator Loss: 0.8756
Epoch 1/5: steps 460 Discriminator Loss: 1.2978... Generator Loss: 0.6877
Epoch 1/5: steps 470 Discriminator Loss: 1.2928... Generator Loss: 0.8425
Epoch 1/5: steps 480 Discriminator Loss: 1.1908... Generator Loss: 0.8765
Epoch 1/5: steps 490 Discriminator Loss: 1.2481... Generator Loss: 0.7307
Epoch 1/5: steps 500 Discriminator Loss: 1.2545... Generator Loss: 0.8466

Epoch 1/5: steps 510 Discriminator Loss: 1.1568... Generator Loss: 0.8647
Epoch 1/5: steps 520 Discriminator Loss: 1.3049... Generator Loss: 0.7341
Epoch 1/5: steps 530 Discriminator Loss: 1.2721... Generator Loss: 0.7575
Epoch 1/5: steps 540 Discriminator Loss: 1.1860... Generator Loss: 0.8390
Epoch 1/5: steps 550 Discriminator Loss: 1.2304... Generator Loss: 0.7792
Epoch 1/5: steps 560 Discriminator Loss: 1.2981... Generator Loss: 0.8205
Epoch 1/5: steps 570 Discriminator Loss: 1.2342... Generator Loss: 0.8837
Epoch 1/5: steps 580 Discriminator Loss: 1.2734... Generator Loss: 0.7386
Epoch 1/5: steps 590 Discriminator Loss: 1.1757... Generator Loss: 0.8840
Epoch 1/5: steps 600 Discriminator Loss: 1.2912... Generator Loss: 0.8048

Epoch 1/5: steps 610 Discriminator Loss: 1.2488... Generator Loss: 0.8480
Epoch 1/5: steps 620 Discriminator Loss: 1.2181... Generator Loss: 0.7336
Epoch 1/5: steps 630 Discriminator Loss: 1.1605... Generator Loss: 0.8206
Epoch 1/5: steps 640 Discriminator Loss: 1.2222... Generator Loss: 0.8090
Epoch 1/5: steps 650 Discriminator Loss: 1.3474... Generator Loss: 0.8090
Epoch 1/5: steps 660 Discriminator Loss: 1.1782... Generator Loss: 1.0533
Epoch 1/5: steps 670 Discriminator Loss: 1.2942... Generator Loss: 0.8093
Epoch 1/5: steps 680 Discriminator Loss: 1.2199... Generator Loss: 0.7276
Epoch 1/5: steps 690 Discriminator Loss: 1.2940... Generator Loss: 0.9284
Epoch 1/5: steps 700 Discriminator Loss: 1.2376... Generator Loss: 0.8444

Epoch 1/5: steps 710 Discriminator Loss: 1.2697... Generator Loss: 0.9803
Epoch 1/5: steps 720 Discriminator Loss: 1.3541... Generator Loss: 0.7483
Epoch 1/5: steps 730 Discriminator Loss: 1.2649... Generator Loss: 0.7835
Epoch 1/5: steps 740 Discriminator Loss: 1.2694... Generator Loss: 0.8413
Epoch 1/5: steps 750 Discriminator Loss: 1.1136... Generator Loss: 0.8834
Epoch 1/5: steps 760 Discriminator Loss: 1.1691... Generator Loss: 0.8551
Epoch 1/5: steps 770 Discriminator Loss: 1.1809... Generator Loss: 0.8820
Epoch 1/5: steps 780 Discriminator Loss: 1.3004... Generator Loss: 0.7246
Epoch 1/5: steps 790 Discriminator Loss: 1.3070... Generator Loss: 1.0823
Epoch 1/5: steps 800 Discriminator Loss: 1.2191... Generator Loss: 0.8287

Epoch 1/5: steps 810 Discriminator Loss: 1.2343... Generator Loss: 0.8341
Epoch 1/5: steps 820 Discriminator Loss: 1.2276... Generator Loss: 0.8424
Epoch 1/5: steps 830 Discriminator Loss: 1.1768... Generator Loss: 0.9659
Epoch 1/5: steps 840 Discriminator Loss: 1.2262... Generator Loss: 0.7477
Epoch 1/5: steps 850 Discriminator Loss: 1.1730... Generator Loss: 1.0589
Epoch 1/5: steps 860 Discriminator Loss: 1.4110... Generator Loss: 0.7671
Epoch 1/5: steps 870 Discriminator Loss: 1.3144... Generator Loss: 0.6294
Epoch 1/5: steps 880 Discriminator Loss: 1.0775... Generator Loss: 0.9440
Epoch 1/5: steps 890 Discriminator Loss: 1.2320... Generator Loss: 0.9406
Epoch 1/5: steps 900 Discriminator Loss: 1.2186... Generator Loss: 0.9377

Epoch 1/5: steps 910 Discriminator Loss: 1.2046... Generator Loss: 0.9343
Epoch 1/5: steps 920 Discriminator Loss: 1.0803... Generator Loss: 1.0031
Epoch 1/5: steps 930 Discriminator Loss: 1.2675... Generator Loss: 0.8125
Epoch 2/5: steps 10 Discriminator Loss: 1.1133... Generator Loss: 1.0187
Epoch 2/5: steps 20 Discriminator Loss: 1.2889... Generator Loss: 0.8226
Epoch 2/5: steps 30 Discriminator Loss: 1.2406... Generator Loss: 0.9133
Epoch 2/5: steps 40 Discriminator Loss: 1.2390... Generator Loss: 0.8919
Epoch 2/5: steps 50 Discriminator Loss: 1.2926... Generator Loss: 0.6769
Epoch 2/5: steps 60 Discriminator Loss: 1.2700... Generator Loss: 0.7776
Epoch 2/5: steps 70 Discriminator Loss: 1.2484... Generator Loss: 0.7927
Epoch 2/5: steps 80 Discriminator Loss: 1.1916... Generator Loss: 0.8733
Epoch 2/5: steps 90 Discriminator Loss: 1.3121... Generator Loss: 0.9195
Epoch 2/5: steps 100 Discriminator Loss: 1.2241... Generator Loss: 0.7955

CelebA¶

Run your GANs on CelebA. It will take around 20 minutes on the average GPU to run one epoch. You can run the whole epoch or stop when it starts to generate realistic faces.

batch_size = 64
z_dim = 100
learning_rate = .0002
beta1 = .5


"""
DON'T MODIFY ANYTHING IN THIS CELL THAT IS BELOW THIS LINE
"""
epochs = 5

celeba_dataset = helper.Dataset('celeba', glob(os.path.join(data_dir, 'img_align_celeba/*.jpg')))
with tf.Graph().as_default():
    train(epochs, batch_size, z_dim, learning_rate, beta1, celeba_dataset.get_batches,
          celeba_dataset.shape, celeba_dataset.image_mode)

Epoch 1/5: steps 10 Discriminator Loss: 1.5192... Generator Loss: 0.6782
Epoch 1/5: steps 20 Discriminator Loss: 1.4686... Generator Loss: 0.7772
Epoch 1/5: steps 30 Discriminator Loss: 1.3323... Generator Loss: 0.7284
Epoch 1/5: steps 40 Discriminator Loss: 1.2378... Generator Loss: 0.9239
Epoch 1/5: steps 50 Discriminator Loss: 1.0628... Generator Loss: 1.0307
Epoch 1/5: steps 60 Discriminator Loss: 0.9794... Generator Loss: 1.0577
Epoch 1/5: steps 70 Discriminator Loss: 0.8427... Generator Loss: 1.2717
Epoch 1/5: steps 80 Discriminator Loss: 0.7993... Generator Loss: 1.3237
Epoch 1/5: steps 90 Discriminator Loss: 0.7460... Generator Loss: 1.1865
Epoch 1/5: steps 100 Discriminator Loss: 0.6681... Generator Loss: 1.5467

Epoch 1/5: steps 110 Discriminator Loss: 0.6745... Generator Loss: 1.6039
Epoch 1/5: steps 120 Discriminator Loss: 0.5455... Generator Loss: 1.8599
Epoch 1/5: steps 130 Discriminator Loss: 0.4885... Generator Loss: 1.7194
Epoch 1/5: steps 140 Discriminator Loss: 0.4652... Generator Loss: 1.9130
Epoch 1/5: steps 150 Discriminator Loss: 0.5569... Generator Loss: 1.9953
Epoch 1/5: steps 160 Discriminator Loss: 0.5278... Generator Loss: 1.8289
Epoch 1/5: steps 170 Discriminator Loss: 0.5327... Generator Loss: 2.1200
Epoch 1/5: steps 180 Discriminator Loss: 0.4889... Generator Loss: 1.8191
Epoch 1/5: steps 190 Discriminator Loss: 0.4039... Generator Loss: 2.4454
Epoch 1/5: steps 200 Discriminator Loss: 0.3689... Generator Loss: 2.4768

Epoch 1/5: steps 210 Discriminator Loss: 0.4191... Generator Loss: 2.7939
Epoch 1/5: steps 220 Discriminator Loss: 0.4046... Generator Loss: 2.3668
Epoch 1/5: steps 230 Discriminator Loss: 0.4185... Generator Loss: 2.4553
Epoch 1/5: steps 240 Discriminator Loss: 0.4496... Generator Loss: 2.4675
Epoch 1/5: steps 250 Discriminator Loss: 0.8382... Generator Loss: 2.0332
Epoch 1/5: steps 260 Discriminator Loss: 0.5890... Generator Loss: 1.3996
Epoch 1/5: steps 270 Discriminator Loss: 0.6819... Generator Loss: 1.3454
Epoch 1/5: steps 280 Discriminator Loss: 1.3955... Generator Loss: 0.4520
Epoch 1/5: steps 290 Discriminator Loss: 0.7761... Generator Loss: 1.3655
Epoch 1/5: steps 300 Discriminator Loss: 0.5360... Generator Loss: 1.5492

Epoch 1/5: steps 310 Discriminator Loss: 0.6289... Generator Loss: 1.9019
Epoch 1/5: steps 320 Discriminator Loss: 0.8050... Generator Loss: 0.9963
Epoch 1/5: steps 330 Discriminator Loss: 0.4926... Generator Loss: 1.8004
Epoch 1/5: steps 340 Discriminator Loss: 0.5318... Generator Loss: 1.6898
Epoch 1/5: steps 350 Discriminator Loss: 0.5728... Generator Loss: 1.7145
Epoch 1/5: steps 360 Discriminator Loss: 0.4023... Generator Loss: 2.4961
Epoch 1/5: steps 370 Discriminator Loss: 0.3638... Generator Loss: 3.3226
Epoch 1/5: steps 380 Discriminator Loss: 0.4047... Generator Loss: 2.9765
Epoch 1/5: steps 390 Discriminator Loss: 0.4360... Generator Loss: 2.0871
Epoch 1/5: steps 400 Discriminator Loss: 0.4058... Generator Loss: 2.7842

Epoch 1/5: steps 410 Discriminator Loss: 0.6540... Generator Loss: 1.3023
Epoch 1/5: steps 420 Discriminator Loss: 0.6973... Generator Loss: 1.0090
Epoch 1/5: steps 430 Discriminator Loss: 0.3561... Generator Loss: 3.6520
Epoch 1/5: steps 440 Discriminator Loss: 0.5642... Generator Loss: 1.6379
Epoch 1/5: steps 450 Discriminator Loss: 0.6127... Generator Loss: 1.2295
Epoch 1/5: steps 460 Discriminator Loss: 0.3779... Generator Loss: 3.1109
Epoch 1/5: steps 470 Discriminator Loss: 0.3635... Generator Loss: 2.1954
Epoch 1/5: steps 480 Discriminator Loss: 0.4375... Generator Loss: 8.0290
Epoch 1/5: steps 490 Discriminator Loss: 0.6376... Generator Loss: 2.7203
Epoch 1/5: steps 500 Discriminator Loss: 0.5318... Generator Loss: 1.4709

Epoch 1/5: steps 510 Discriminator Loss: 0.4356... Generator Loss: 3.1780
Epoch 1/5: steps 520 Discriminator Loss: 0.4738... Generator Loss: 1.8482
Epoch 1/5: steps 530 Discriminator Loss: 0.3973... Generator Loss: 2.5151
Epoch 1/5: steps 540 Discriminator Loss: 0.5138... Generator Loss: 1.6781
Epoch 1/5: steps 550 Discriminator Loss: 0.3654... Generator Loss: 3.0520
Epoch 1/5: steps 560 Discriminator Loss: 0.3752... Generator Loss: 3.3289
Epoch 1/5: steps 570 Discriminator Loss: 0.5028... Generator Loss: 1.4758
Epoch 1/5: steps 580 Discriminator Loss: 0.4306... Generator Loss: 3.5621
Epoch 1/5: steps 590 Discriminator Loss: 0.3273... Generator Loss: 5.0551
Epoch 1/5: steps 600 Discriminator Loss: 0.3590... Generator Loss: 3.9476

Epoch 1/5: steps 610 Discriminator Loss: 0.4265... Generator Loss: 2.5437
Epoch 1/5: steps 620 Discriminator Loss: 0.4193... Generator Loss: 4.0507
Epoch 1/5: steps 630 Discriminator Loss: 0.3707... Generator Loss: 2.8156
Epoch 1/5: steps 640 Discriminator Loss: 0.4921... Generator Loss: 3.3273
Epoch 1/5: steps 650 Discriminator Loss: 0.3383... Generator Loss: 4.1146
Epoch 1/5: steps 660 Discriminator Loss: 0.4943... Generator Loss: 1.8476
Epoch 1/5: steps 670 Discriminator Loss: 0.3974... Generator Loss: 2.9515
Epoch 1/5: steps 680 Discriminator Loss: 0.3838... Generator Loss: 2.2258
Epoch 1/5: steps 690 Discriminator Loss: 0.3646... Generator Loss: 2.6790
Epoch 1/5: steps 700 Discriminator Loss: 0.3603... Generator Loss: 2.6738

Epoch 1/5: steps 710 Discriminator Loss: 0.3577... Generator Loss: 2.8696
Epoch 1/5: steps 720 Discriminator Loss: 0.3549... Generator Loss: 2.8707
Epoch 1/5: steps 730 Discriminator Loss: 0.3916... Generator Loss: 2.1700
Epoch 1/5: steps 740 Discriminator Loss: 0.3937... Generator Loss: 2.7461
Epoch 1/5: steps 750 Discriminator Loss: 0.6986... Generator Loss: 1.1114
Epoch 1/5: steps 760 Discriminator Loss: 0.6450... Generator Loss: 1.2556
Epoch 1/5: steps 770 Discriminator Loss: 0.3840... Generator Loss: 5.2096
Epoch 1/5: steps 780 Discriminator Loss: 0.3124... Generator Loss: 7.0366
Epoch 1/5: steps 790 Discriminator Loss: 0.5434... Generator Loss: 3.4836
Epoch 1/5: steps 800 Discriminator Loss: 0.5097... Generator Loss: 1.9451

Epoch 1/5: steps 810 Discriminator Loss: 0.4832... Generator Loss: 1.7235
Epoch 1/5: steps 820 Discriminator Loss: 0.4093... Generator Loss: 2.0122
Epoch 1/5: steps 830 Discriminator Loss: 0.4718... Generator Loss: 2.3181
Epoch 1/5: steps 840 Discriminator Loss: 0.3593... Generator Loss: 2.3013
Epoch 1/5: steps 850 Discriminator Loss: 0.3818... Generator Loss: 2.5707
Epoch 1/5: steps 860 Discriminator Loss: 0.3997... Generator Loss: 2.2496
Epoch 1/5: steps 870 Discriminator Loss: 0.3905... Generator Loss: 2.2576
Epoch 1/5: steps 880 Discriminator Loss: 0.3372... Generator Loss: 3.9857
Epoch 1/5: steps 890 Discriminator Loss: 0.4761... Generator Loss: 1.6131
Epoch 1/5: steps 900 Discriminator Loss: 0.7294... Generator Loss: 1.4536

Epoch 1/5: steps 910 Discriminator Loss: 0.3868... Generator Loss: 2.7961
Epoch 1/5: steps 920 Discriminator Loss: 0.8857... Generator Loss: 3.3619
Epoch 1/5: steps 930 Discriminator Loss: 0.5110... Generator Loss: 1.5866
Epoch 1/5: steps 940 Discriminator Loss: 0.5366... Generator Loss: 1.8076
Epoch 1/5: steps 950 Discriminator Loss: 0.5036... Generator Loss: 2.5187
Epoch 1/5: steps 960 Discriminator Loss: 0.3606... Generator Loss: 3.4227
Epoch 1/5: steps 970 Discriminator Loss: 0.3711... Generator Loss: 3.2038
Epoch 1/5: steps 980 Discriminator Loss: 0.5526... Generator Loss: 1.9197
Epoch 1/5: steps 990 Discriminator Loss: 1.1375... Generator Loss: 1.0391
Epoch 1/5: steps 1000 Discriminator Loss: 0.4423... Generator Loss: 2.6213

Submitting This Project¶

When submitting this project, make sure to run all the cells before saving the notebook. Save the notebook file as "dlnd_face_generation.ipynb" and save it as a HTML file under "File" -> "Download as". Include the "helper.py" and "problem_unittests.py" files in your submission.